TRIBUNAL DE CONTAS DA UNIO SecretariaGeral de Controle
TRIBUNAL DE CONTAS DA UNIÃO Secretaria-Geral de Controle Externo Secretaria de Controle Externo da Previdência, do Trabalho e da Assistência Social Continuous Supervision on Social Benefits
Structure of the presentation • Context • What is TCU? • Methodology: • Government Auditing Generations • Cycle 2015 – Results: • Cycle 2016: • Pensions • Social Assistance • Labor
Context TCU President’s Office Congress Ministries TCU (Supreme Audit Institution) Supreme Court
Context Annual Budget & Numbers • Pensions • USD 138 bi/yr • +30 mi pymt/mo • Labor • USD 16. 7 bi • +600 k pymt/2 wk • Social Assistance • USD 18. 1 bi • Cad. Unico: 120 mi • Bolsa Familia: 25 mi pymt/mo • Debt related expenditure: USD 328 bi • Total: USD 590 bi 1. 00 R$ = US$ 0. 26 (12/31/2015 PTAX/BACEN)
Methodology Previous works • Continuous Auditing x Continuous Supervision • Types of Government Auditing: Compliance, Performance, Financial • Evolution of techniques in Governmental Data Auditing of Social Benefits: • 1 st Generation (until 2006) • Traditional auditing of physical documents • 2 nd Generation (2006 -2015) • Moderate use of IT tools (crossing of few databases) • 3 rd Generation (2015 -2016) • Automatization of procedures (increase in volume of data) • 4 th Generation (2016 -) • Use of statistical and machine learning techniques (increase in volume and complexity of data and new expected products)
Methodology Continuous Supervision • Credibility • Public databases show high rate of missing values and contradictory registration data • Importance of Data Quality techniques and Credibility indicators • Timing of auditing • Massive amount of payments (pensions and social assistance benefits) and short cycles of benefit granting (unemployment benefits) • Few audits per year vs. Continuous monthly (weekly) checks • Fraud complexity • Use of forged documents • Organized groups of fraudsters • Patterns of fraud
Methodology Fluxogram
Cycle 2015 Results – Pensions • Implementation of 4 non-compliance typologies • Undue accumulation of benefits • Benefit with NIT (worker ID number) shared by more than one holder • Benefit value over maximum allowed • Rural benefit unduly payed to urban worker • Possible yearly savings of USD 48 mi • Moderate use of Data Quality and Automation tools • Implementation of a Pension Panel (BI tool for auditing) 1. 00 R$ = US$ 0. 26 (12/31/2015 PTAX/BACEN)
Cycle 2015 Results – Labor • Focus on Data Collection and ETL processes • Implementation of 4 non-compliance typologies • Unemployment benefits for Artisanal Fishermen with incompatible income from other sources (private sector) • Unemployment benefits for Artisanal Fishermen with incompatible income from other sources (Pension benefits) • Unemployment benefits for Artisanal Fishermen with incompatible income from other sources (Brazilian Conditional Cash Transfer Program – Bolsa Familia) • Unemployment benefits for Artisanal Fishermen to deceased workers • Possible savings of USD 1 mi (R$ 4 mi) 1. 00 R$ = US$ 0. 26 (12/31/2015 PTAX/BACEN)
Cycle 2015 Results – Social Assistance • Implementation of 1 non-compliance typology • Bolsa Familia program benefits with incompatible income from other sources (data crossing) • Possible yearly savings of USD 50 mi • Data Quality test on the Cad. Unico (Social Assistance database) 19, 2% 74, 0% 26, 0% 6, 7% 0, 1% Match exato Combinado Indeterminado 1. 00 R$ = US$ 0. 26 (12/31/2015 PTAX/BACEN) Não combinado 56 k irregular cases
Cycle 2016 Pensions • Data Mining in Pension Fraud/Irregularities • Target universe • 4, 359 frauded benefits • 21, 789 irregular benefits 33 mi total of population • 242, 927 regular benefits • Time frame • Aug – Dec • Current phase: Modeling • Preliminary results • “Flat table” with 3, 203 indicators • 242, 927+4, 359 (fraud) / 242, 927+21, 789 (irregular) rows • Calculated by standard-deviations of subgroup (dimension x metric x unit of observation)
Cycle 2016 Pensions • Data Mining in Pension Fraud/Irregularities • Main challenges • Lack of information on old frauded benefits • Available data: jan/2014 – aug/2016 • Original fraud cases of 29. 538 reduced to 4. 359 • Poor Data Quality (high missing ratio on several columns)
Cycle 2016 Labor • Automation of Data Collection, ETL and 4 previous typologies • Implementation of illegal accumulation for Unemployment benefits (formal worker) • Artisanal Fishermen: 363 pymt (jan/16) • Concentrated in time (depending on the species) and region • Formal worker: 2 mi pymt (jan/16) • Up to 5 payments, nationwide
Cycle 2016 Social Assistance • Data Quality/Enrichment of Cad. Unico • Electoral database (high quality, biometry) • Private sector employment information • Implementation of typologies • Incompatibility between self-declaratory low income and wealth (societal participation, expensive vehicles) • Bolsa Familia program benefits with incompatible income from Government sources (pensions, unemployment, public contracts)
• Contacts: • Rodrigo Hildebrand • NCAD – Center of Data Analysis and Information Technology • Department of External Control – Social Security, Labor and Welfare • rodrigooc@tcu. gov. br – +55 (61) 3316 -7971 • secexprevi@tcu. gov. br – +55 (61) 3316 -7365 Thank you!!
- Slides: 15